Smart Road Studs With Magnetic Sensors for Multilane Traffic Volume Detection

Abstract

Traffic detection is essential in intelligent transportation systems. Magnetic sensors, valued for their compactness, low cost, and robustness to interference, show promise as traffic detectors. Yet their current vehicle detection abilities fall short of widespread deployment. This study addresses this gap by integrating magnetic sensors into smart road studs (SRSs), enabling adaptive sensing and control capabilities. Conventional detection algorithms for roadside or center-lane placement are unsuitable for sensors on lane markings. This article introduces a multilane traffic volume detection algorithm tailored for the SRS network. A multiscale convolutional neural network (MSCNN) module based on 1-D convolution is first designed to automatically extract multiscale features from individual signals. Then, the C-Transformer (C-Trans) and S-Transformer (S-Trans) encoding modules, built on Transformer architecture, are employed to capture both intrasignal and spatial intersensor correlations. By merging multiscale, correlation-based, and manually extracted features, the proposed method facilitates multilane vehicle detection. To further refine accuracy and underscore the role of key sensor nodes, a single-sensor vehicle detection approach is integrated, with the Dempster-Shafer (D-S) theory used for result fusion. Experimental results demonstrate that the proposed approach achieves multilane traffic volume detection with an error rate of approximately 1.6%, outperforming current methods.